This is nothing to do with ActivityPub. It’s to do with Mastodon’s custom implementation of “private” posts.
Making it extremely clear to everyone that random server software can expose Mastodon’s “private” posts is absolutely the right way to handle disclosure here. Dan didn’t even try to do that, he just fixed the bug, but if he had made a big post saying “hey this is not my fault Mastodon private posts are not private, here’s full explanation about what’s going on” I think that would have been completely fine. This is not a “vulnerability” in the traditional sense like a buffer overflow, it’s just a design flaw in Mastodon which other softwares are by convention agreeing to cater to. I think the culture of security (and the level of clue in general) in the Fediverse has wandered into territory where “let’s all pretend that these posts are secure and get mad at anyone who reveals that they are not” is widely accepted now, but that doesn’t make it right.
With the full disclosure approach, the full details of the vulnerability are made public as soon as they are identified. This means that the full details (sometimes including exploit code) are available to attackers, often before a patch is available. The full disclosure approach is primarily used in response or organizations ignoring reported vulnerabilities, in order to put pressure on them to develop and publish a fix.
This makes the full disclosure approach very controversial, and it is seen as irresponsible by many people. Generally it should only be considered as a last resort, when all other methods have failed, or when exploit code is already publicly available.
Responsible or Coordinated Disclosure
Responsible disclosure attempts to find a reasonable middle ground between these two approaches. With responsible disclosure, the initial report is made privately, but with the full details being published once a patch has been made available (sometimes with a delay to allow more time for the patches to be installed).
Hm… maybe. The exact nature of the problem in Pixelfed means that anyone with a Pixelfed account on a server which is getting private statuses can choose to follow someone who’s set to “approve followers” and then read all the private statuses. I do see how that’s significantly worse than just the normal lay-of-the-land of the problem, which is a little more random, and laying that out as a little roadmap to read someone else’s private statuses before there’s been a nice responsible length of time for things to get fixed could be seen as worsening the problem.
The point that I’m making is that anyone who’s posting private statuses to Mastodon and expecting them to stay private is making a bad mistake already. The structure of the protocol is such that they can’t be assured of staying private regardless of what Pixelfed did or even if Pixelfed didn’t exist. They’re getting federated to servers whose behavior is not assured, in a way where a conformant ActivityPub implementation can expose them. People who are posting private statuses need to understand that.
That whole blog post where the person is talking about her partner writing private statuses, and then the gut-wrenching realization that they were being exposed on Pixelfed… but then the resolution being “Pixelfed fucked up I hate Dansup now” and then continuing to post the private statuses, is wrong. That person’s partner needs to stop treating their private posts on Mastodon that way. The timer for responsible disclosure started circa 2017 or whenever Mastodon decided on how to implement their private statuses. It’s been and gone.
Like I say, I get the harm-reduction aspect of saying it would have been better if Dansup was a little more discreet about this particularly bad attack vector until a few more days went by for everyone to upgrade. But it hardly matters. There are still server softwares our there that are going to be exposing people’s private Mastodon posts. It’s just how federation between untrusted servers works. Giving people the illusion that if Dan had just been more discreet then this harm would have been reduced is lulling them into a false sense of security, in my view.
Are you hoping to restart our disagreement through sheer passive-aggressiveness? Okay, sure.
In my view, this is a Mastodon design flaw (or a user-expectation issue or whatever you want to call it.) I already said that, and you’re involved in the unproductive-arguer’s pastime of pretending not to understand that that’s my position, and just aggressively repeatedly reframing things according to your position and hoping I’ll knuckle under to it through sheer force of repetition.
I’m not super invested in trying to track down each and every software that might manage to expose the “private” statuses in this way. I just know that as things come and go there are guaranteed to be some. If you have an mbin account and Mastodon account, though, we can try a little experiment. I don’t know the outcome, I’m just curious after taking a quick look down the FediDB list and a quick grep through mbin’s source code. You can be the one to responsibly disclose to mbin how their ActivityPub-conforming behavior is a problem, if indeed it turns out that it is, since you seem to be extremely committed to the idea that the model of “vulnerability” needs to be applied to this particular ActivityPub-conforming behavior. Since you’re a security researcher, having that as a CVE you discovered can be an achievement for you. It’s all yours, you can have it.
Because it is transparently obvious that it’s going to happen.
If you’re sending your users’ private statuses to an ActivityPub server, and just hoping that it’s going to choose to keep them private according to certain parameters even though that’s not what the spec stays it needs to do, then you’re fucking up. The fact that we know that particular instances of particular software are exposing them is a nice demonstration of the harm, a confirmation that you’re fucking up when you’re doing that, but it’s not really needed. It is the absolutely predictable result of some basic principles of security which, as a security researcher, you should absolutely be aware of.
I’ve repeatedly explained this. You’ve repeatedly explained your position. We’ve both had our say. You seem addicted to the concept of “winning” the conversation and wanting to just go back and forth. In that case I would really encourage you to state your position again, and I can state mine again, and we can both have fun doing that for a while. Want to? It sounds like a productive use of both of our time. It’s fun, too.
Edit: Actually, I didn’t even realize you are on fedia.io when I was typing this. You can test for yourself whether mbin does this, too, by coordinating with @[email protected]. Follow his user, then have him post one of those private statuses, then fetch his user profile via fedia.io from an incognito window and see whether the private statuses show up. I have no idea whether they will, but if I had to guess, I would say it’s better than even odds.
It has everything to do with ActivityPub since if you follow that protocol strictly you will cause this behavior. It still doesn’t change that Dansup was told that this caused Bad Things™ and yet he didn’t follow normal procedure in how you handle it.
Vulnerabilities don’t need to be buffer overflows.
It has everything to do with ActivityPub since if you follow that protocol strictly you will cause this behavior.
Absolutely not. Which part of the spec? I linked up there to quite a thorough explanation of what the spec does and doesn’t dictate in this area, and how Mastodon chooses to behave in its implementation. What part of my explanation did I get wrong? Are they violating 5.1, 5.2, 7.1, some other part? How?
/cybersec researcher
I do not believe you. “I’m sending things out which need to be handled carefully in a protocol-nonstandard way by the recipient server software (which could be literally anything), or else my user’s private posts will be exposed. If someone talks about that situation and lets people know what’s going on, that’s irresponsible disclosure.”
If you actually are a cybersec researcher, you are bad at your job.
I’ve said nothing about any spec violation. That’s not relevant.
It has everything to do with ActivityPub since if you follow that protocol strictly you will cause this behavior.
That’s what I was going by. I guess I could re-read this now and interpret “this behavior” as Pixelfed’s side, instead of Mastodon’s side as I initially read it, and decide that you are agreeing with me that Mastodon’s behavior was (and is) out of spec? Do I have that right?
It still doesn’t change that Dansup was told that this caused Bad Things™ and yet he didn’t follow normal procedure in how you handle it.
It is normal procedure to fix a bug when you are notified about it.
The design flaw in Mastodon that managed to bite Pixelfed in this situation still exists. People were writing about it back in 2017 when this was all being first implemented. The idea that “normal procedure” needs to include keeping it a secret that Mastodon’s “private” statuses can be exposed by any server software that doesn’t handle them in the way that’s expected, is 100% wrong.
I’ll rephrase what I said earlier: Since you’re a security researcher, and you apparently think Dan should have played into the idea of keeping it a secret that Mastodon’s private statuses are not secret by obfuscating the information about how he was fixing Pixelfed to more effectively hide them, you are bad at your job. In this instance. The fault lies with how private statuses are implemented, and nothing about that needs to be kept secret as would a normal vulnerability, during responsible disclosure. In fact, it is extremely harmful to let users believe that these privacy settings are anything other than vague recommendations, specifically because of the risk they will act accordingly and expose some of their private posts to the world. They should know exactly what’s going on with it, and Dan accidentally failing to keep that a secret is in no way causing bad things.
You have absolutely no idea what “responsible” in “responsible disclosure” means :) It’s completely irrelevant how Mastodon has implemented private posts when it comes to how Dansup handled the issue, knowing what the effects were.
You don’t, when told of a vulnerability, handle it in a way that cause harm if it can be avoided.
Yeah, you said that stuff before and then you said it again. I do understand what your argument is here. I was trying a couple of different ways of explaining what I was saying in response, but it seems like it’s not working. Oh well.
Maybe I’m wrong, but shouldn’t posts only be insecure if they’re propagated to the insecure instance?
Is any private post visible to people on servers that the poster doesn’t have followers on?
Could I curl the uri of a post thats “private” and get the post’s content?
Maybe I’m wrong, but shouldn’t posts only be insecure if they’re propagated to the insecure instance?
“Insecure” in this case simply means any server that doesn’t implement Mastodon’s custom handling for “private” posts. With that definition, the answer to your question is yes. It has been mentioned by Mastodon people that this is a significant problem for the ability to actually keep these private posts private in the real world. The chance of it going wrong is small (depending on your follower count) but the potential for harm is very large. I would therefore go further, and say that it’s a very bad thing that Mastodon is telling people that these posts are “private” when the mechanism which is supposed to keep them private is so unreliable.
Is any private post visible to people on servers that the poster doesn’t have followers on?
It is not. If you’re sufficiently careful with approving your followers, making sure that each of them is on an instance that’s going to handle private posts the way you expect, then you’re probably fine.
Could I curl the uri of a post thats “private” and get the post’s content?
If it’s been federated to an insecure server then yes. If not then I think no.
Mastodon really is the internet explorer of the fediverse.
In any case, I don’t think its that bad. I would compare it to an email provider accidentially leaking messages. Still bad, but its not a reason to abandon email as a means of communication.
We should encrypt posts, like diaspora does. Like how we should pgp encrypt emails, but no one will.
also, I just checked myself, a random “private” post I made isn’t accessible over AP if I curl it unauthenticated.
Running curl.exe https://calckey.world/notes/a63slz8j6l -H "Accept: application/activity+json" returns nothing, but replacing the uri with a public post does show it.
An insecure server’s copy of the post isn’t accessible over AP, only the original post’s link should return anything.
I usually agree with you, but here @[email protected] is right.
https://cheatsheetseries.owasp.org/cheatsheets/Vulnerability_Disclosure_Cheat_Sheet.html#full-disclosure
Hm… maybe. The exact nature of the problem in Pixelfed means that anyone with a Pixelfed account on a server which is getting private statuses can choose to follow someone who’s set to “approve followers” and then read all the private statuses. I do see how that’s significantly worse than just the normal lay-of-the-land of the problem, which is a little more random, and laying that out as a little roadmap to read someone else’s private statuses before there’s been a nice responsible length of time for things to get fixed could be seen as worsening the problem.
The point that I’m making is that anyone who’s posting private statuses to Mastodon and expecting them to stay private is making a bad mistake already. The structure of the protocol is such that they can’t be assured of staying private regardless of what Pixelfed did or even if Pixelfed didn’t exist. They’re getting federated to servers whose behavior is not assured, in a way where a conformant ActivityPub implementation can expose them. People who are posting private statuses need to understand that.
That whole blog post where the person is talking about her partner writing private statuses, and then the gut-wrenching realization that they were being exposed on Pixelfed… but then the resolution being “Pixelfed fucked up I hate Dansup now” and then continuing to post the private statuses, is wrong. That person’s partner needs to stop treating their private posts on Mastodon that way. The timer for responsible disclosure started circa 2017 or whenever Mastodon decided on how to implement their private statuses. It’s been and gone.
Like I say, I get the harm-reduction aspect of saying it would have been better if Dansup was a little more discreet about this particularly bad attack vector until a few more days went by for everyone to upgrade. But it hardly matters. There are still server softwares our there that are going to be exposing people’s private Mastodon posts. It’s just how federation between untrusted servers works. Giving people the illusion that if Dan had just been more discreet then this harm would have been reduced is lulling them into a false sense of security, in my view.
If you know of other ActivityPub servers that expose private posts the same way I suggest you make a responsible disclosure to the developers.
I don’t know of any, but you claim they exist so …
Are you hoping to restart our disagreement through sheer passive-aggressiveness? Okay, sure.
In my view, this is a Mastodon design flaw (or a user-expectation issue or whatever you want to call it.) I already said that, and you’re involved in the unproductive-arguer’s pastime of pretending not to understand that that’s my position, and just aggressively repeatedly reframing things according to your position and hoping I’ll knuckle under to it through sheer force of repetition.
I’m not super invested in trying to track down each and every software that might manage to expose the “private” statuses in this way. I just know that as things come and go there are guaranteed to be some. If you have an mbin account and Mastodon account, though, we can try a little experiment. I don’t know the outcome, I’m just curious after taking a quick look down the FediDB list and a quick grep through mbin’s source code. You can be the one to responsibly disclose to mbin how their ActivityPub-conforming behavior is a problem, if indeed it turns out that it is, since you seem to be extremely committed to the idea that the model of “vulnerability” needs to be applied to this particular ActivityPub-conforming behavior. Since you’re a security researcher, having that as a CVE you discovered can be an achievement for you. It’s all yours, you can have it.
You could’ve saved yourself a lot of typing there by just admitting to claiming things you actually didn’t know.
Because it is transparently obvious that it’s going to happen.
If you’re sending your users’ private statuses to an ActivityPub server, and just hoping that it’s going to choose to keep them private according to certain parameters even though that’s not what the spec stays it needs to do, then you’re fucking up. The fact that we know that particular instances of particular software are exposing them is a nice demonstration of the harm, a confirmation that you’re fucking up when you’re doing that, but it’s not really needed. It is the absolutely predictable result of some basic principles of security which, as a security researcher, you should absolutely be aware of.
I’ve repeatedly explained this. You’ve repeatedly explained your position. We’ve both had our say. You seem addicted to the concept of “winning” the conversation and wanting to just go back and forth. In that case I would really encourage you to state your position again, and I can state mine again, and we can both have fun doing that for a while. Want to? It sounds like a productive use of both of our time. It’s fun, too.
Edit: Actually, I didn’t even realize you are on fedia.io when I was typing this. You can test for yourself whether mbin does this, too, by coordinating with @[email protected]. Follow his user, then have him post one of those private statuses, then fetch his user profile via fedia.io from an incognito window and see whether the private statuses show up. I have no idea whether they will, but if I had to guess, I would say it’s better than even odds.
It has everything to do with ActivityPub since if you follow that protocol strictly you will cause this behavior. It still doesn’t change that Dansup was told that this caused Bad Things™ and yet he didn’t follow normal procedure in how you handle it.
Vulnerabilities don’t need to be buffer overflows.
/cybersec researcher
Absolutely not. Which part of the spec? I linked up there to quite a thorough explanation of what the spec does and doesn’t dictate in this area, and how Mastodon chooses to behave in its implementation. What part of my explanation did I get wrong? Are they violating 5.1, 5.2, 7.1, some other part? How?
I do not believe you. “I’m sending things out which need to be handled carefully in a protocol-nonstandard way by the recipient server software (which could be literally anything), or else my user’s private posts will be exposed. If someone talks about that situation and lets people know what’s going on, that’s irresponsible disclosure.”
If you actually are a cybersec researcher, you are bad at your job.
hahahahaha
Watch and try again ;) I post under my real name.
https://www.cve.org/CVERecord?id=CVE-2024-44754
https://www.youtube.com/watch?v=ZbKLAjPYOEg
Feel free to post less and read more.
Okay. What part of the spec did Pixelfed violate? Where in the spec is Mastodon’s implementation of private posts justified?
Read more, post less. I’ve said nothing about any spec violation. That’s not relevant.
That’s what I was going by. I guess I could re-read this now and interpret “this behavior” as Pixelfed’s side, instead of Mastodon’s side as I initially read it, and decide that you are agreeing with me that Mastodon’s behavior was (and is) out of spec? Do I have that right?
It is normal procedure to fix a bug when you are notified about it.
The design flaw in Mastodon that managed to bite Pixelfed in this situation still exists. People were writing about it back in 2017 when this was all being first implemented. The idea that “normal procedure” needs to include keeping it a secret that Mastodon’s “private” statuses can be exposed by any server software that doesn’t handle them in the way that’s expected, is 100% wrong.
I’ll rephrase what I said earlier: Since you’re a security researcher, and you apparently think Dan should have played into the idea of keeping it a secret that Mastodon’s private statuses are not secret by obfuscating the information about how he was fixing Pixelfed to more effectively hide them, you are bad at your job. In this instance. The fault lies with how private statuses are implemented, and nothing about that needs to be kept secret as would a normal vulnerability, during responsible disclosure. In fact, it is extremely harmful to let users believe that these privacy settings are anything other than vague recommendations, specifically because of the risk they will act accordingly and expose some of their private posts to the world. They should know exactly what’s going on with it, and Dan accidentally failing to keep that a secret is in no way causing bad things.
You have absolutely no idea what “responsible” in “responsible disclosure” means :) It’s completely irrelevant how Mastodon has implemented private posts when it comes to how Dansup handled the issue, knowing what the effects were.
You don’t, when told of a vulnerability, handle it in a way that cause harm if it can be avoided.
Yeah, you said that stuff before and then you said it again. I do understand what your argument is here. I was trying a couple of different ways of explaining what I was saying in response, but it seems like it’s not working. Oh well.
Maybe I’m wrong, but shouldn’t posts only be insecure if they’re propagated to the insecure instance? Is any private post visible to people on servers that the poster doesn’t have followers on?
Could I
curlthe uri of a post thats “private” and get the post’s content?“Insecure” in this case simply means any server that doesn’t implement Mastodon’s custom handling for “private” posts. With that definition, the answer to your question is yes. It has been mentioned by Mastodon people that this is a significant problem for the ability to actually keep these private posts private in the real world. The chance of it going wrong is small (depending on your follower count) but the potential for harm is very large. I would therefore go further, and say that it’s a very bad thing that Mastodon is telling people that these posts are “private” when the mechanism which is supposed to keep them private is so unreliable.
https://marrus-sh.github.io/mastodon-info/everything-you-need-to-know-about-privacy-v1.3-020170427.html
https://github.com/mastodon/mastodon/issues/712
It is not. If you’re sufficiently careful with approving your followers, making sure that each of them is on an instance that’s going to handle private posts the way you expect, then you’re probably fine.
If it’s been federated to an insecure server then yes. If not then I think no.
Mastodon really is the internet explorer of the fediverse.
In any case, I don’t think its that bad. I would compare it to an email provider accidentially leaking messages. Still bad, but its not a reason to abandon email as a means of communication.
We should encrypt posts, like diaspora does. Like how we should pgp encrypt emails, but no one will.
also, I just checked myself, a random “private” post I made isn’t accessible over AP if I curl it unauthenticated. Running
curl.exe https://calckey.world/notes/a63slz8j6l -H "Accept: application/activity+json"returns nothing, but replacing the uri with a public post does show it.An insecure server’s copy of the post isn’t accessible over AP, only the original post’s link should return anything.